Сальников Александр Святославович
Analysis of Cryptocurrency Transaction Data Using Graph Structural Information
Науки о данных
Distributed Ledger is an owner-less database consists of transactions between network participants. That data can be easily interpreted as a large graph of interactions between entities that contains a lot of node structural information and metadata. Which makes it a good target of applying Machine Learning algorithms, particularly clustering. In this paper we consider node2vec - modern method of embedding of graph node’s structural information into low-dimensional vector space. We conduct a comparative study of pure graph based clustering algorithms with node2vec + K-means conjunction and show that the latter is superior not only in terms of accuracy of extracting clusters from SBM Random Graph but also in terms of scalability. Finally, node2vec embeddings for a subset of Ethereum Transactions data were generated. Visual representation of low-dimension vectors is given and clustering performed.